Online images aren't keeping up when it comes to addressing gender bias

Embargoed until: Publicly released: 2024-02-15 03:00

International

Ella Maru

As we spend more time consuming images online and less time reading, international researchers say we could be exposing ourselves to greater gender bias. The team collected over a million images and billions of words across Google, Wikipedia and IMDb and analysed them for gender biases around which genders were represented in about 3,500 social categories such as 'nurse' or 'banker'. The researchers say the images were consistently more likely to overrepresent men or women in social categories with gender stereotypes. They then conducted an experiment where 450 people were asked to search either Google News or Google images for specific occupations and then report what gender they associated with each occupation - and the researchers say those searching for images were more likely to report gender biases.

News release

From: Springer Nature

Social sciences: Images paint a picture of gender bias *IMAGE* (N&V)

Online images show a stronger gender bias than online texts, according to a study in Nature this week. Exposure to such visual gender stereotypes seems to influence gender biases in people’s beliefs, the findings suggest.

Image consumption is increasing, be it through social media platforms, news agencies or digital advertisers. Such increases may be driven in part by the fact that people process images more quickly, implicitly and memorably than text. Research suggests that images may be more prone to displaying gender bias than text, as text can be made gender neutral, but large-scale analyses are lacking.

To examine the extent of gender bias in online images, and compare it with that in text, Douglas Guilbeault and colleagues assess over one million images from Google, Wikipedia and IMDb, as well as texts from these platforms. They probe gender biases associated with nearly 3,500 social categories, including professions such as doctor and lawyer, or social roles such as neighbour, friend and colleague. For the images, faces were classified by gender (2% were classified as non-binary but were excluded as the analysis focused on perceived gender). Males were over-represented in the search results, but more so in images than texts: 56% of categories are male-skewed according to texts from Google News compared with 62% according to Google Images. For example, searches for plumber, police chief and carpenter were more likely to show male faces, whereas ballet dancer, hairstylist and nurse were more likely to show female faces.

Guilbeault and colleagues go on to look at the implications of these biases with an experiment in which 450 individuals were asked to search for specific occupations in either Google News or Google images; a third control group searched for random objects. Each participant was then asked to rate which gender they most associate with specific occupations. The authors find that people that searched for images exhibited stronger gender bias compared with those who searched for text and the control group, and these biases endure for a few days.

Addressing gender biases in online images will be essential for making the internet a fair and inclusive place, the authors conclude.

Attachments

Note: Not all attachments are visible to the general public. Research URLs will go live after the embargo ends.

Research Springer Nature, Web page The URL will go live after the embargo ends

Journal/
conference: Nature

Research:Paper

Organisation/s: University of California, USA

Funder: We acknowledge members of the MRL (Macro Research Lunch) seminar at the Haas School of Business and the Computational Culture lab, as well as Nina Guilbeault, Nicholas Guilbeault, P. Reginato and R. Lo Sardo for helpful feedback on this project. We thank S. Nanniyur for assistance with data collection. This project was funded by grants from the Fisher Center for Business Analytics; the Center for Equity, Gender and Leadership; and the Barbara and Gerson Bakar Fellowship, each awarded to D.G., and by a grant from the Cora Jane Flood Endowment Fund awarded to S.D., all through the University of California, Berkeley.

Media Contact/s

Contact details are only visible to registered journalists.